We introduce image retrieval methods based on the strokes produced by stochastic paintbrush rendering of image(SPT). SPT simulates the painting process with sequential strokes at decreasing scale of brush-sizes. Features extracted from strokes represent not only the painting's information in the color space, but in the spacial space as well. These methods are tested on a image database containing 510 images of various kinds.
Zoltan Kato OR Ji Xiaowen